An optimized extreme learning machine composite framework for point, probabilistic, and quantile regression forecasting of carbon price

作     者：Wang, Xuming Zhou, Jiaqi Yu, Xiaobing 

作者机构：Nanjing Univ Informat Sci & Technol Engn Training Ctr Nanjing 210044 Peoples R China Nanjing Univ Informat Sci & Technol Sch Management Sci & Engn Nanjing 210044 Peoples R China 

出 版 物：《PROCESS SAFETY AND ENVIRONMENTAL PROTECTION》 (Process Saf. Environ. Prot.)

年 卷 期：2025年第195卷

核心收录：

学科分类：0830[工学-环境科学与工程（可授工学、理学、农学学位）] 0817[工学-化学工程与技术] 08[工学] 

基　　金：Humanities and Social Sciences Fund of the Ministry of Education, China [22YJC630144] NUIST Students' Platform for Innovation and Entrepreneurship Training Program [XJDC202310300092] 

主　　题：Carbon emission trading price Time series forecasting Point forecasting Probabilistic forecasting Interval forecasting 

摘      要：The increasing impact of carbon dioxide emissions on the ecological environment has aroused widespread concern. Precise estimation of carbon prices is essential for rational planning of carbon emissions and mitigation of environmental crises. Carbon price series exhibit nonlinear features, and capturing these features is challenging. Therefore, this paper proposes a carbon price forecasting framework (EEMD-ELM-EGWO) based on signal decomposition and Extreme Learning Machine (ELM). Under this forecasting framework, we realize point forecasting, probabilistic forecasting, and quantile regression interval forecasting. Firstly, the carbon price is decomposed into multiple Intrinsic Mode Functions (IMFs) using Ensemble Empirical Mode Decomposition (EEMD). Subsequently, an intelligently optimized ELM is used to forecast each IMF. Finally, the individual forecasts are aggregated to produce the final forecast. To optimize the hidden layer matrices of the ELM, an Enhanced Grey Wolf Optimizer (EGWO) algorithm is developed. Experimental results show that the proposed method provides more accurate point forecasting results and exhibits superior nonlinear fitting ability compared to other signal processing techniques and intelligent algorithms. In addition, the proposed framework is able to more accurately characterize the uncertainty of carbon price series. Together, these three forecasting models provide carbon market participants with more comprehensive and accurate information on future price trends, which helps to formulate rational policy planning and market strategies.